Download Randomized Controlled Trials (Experimental Studies)

yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Document related concepts

Public health genomics wikipedia, lookup

Declaration of Helsinki wikipedia, lookup

Race and health wikipedia, lookup

Transtheoretical model wikipedia, lookup

Forensic epidemiology wikipedia, lookup

Randomized controlled trial wikipedia, lookup

Clinical trial wikipedia, lookup

Management of multiple sclerosis wikipedia, lookup

Human subject research wikipedia, lookup

Epidemiology wikipedia, lookup

Multiple sclerosis research wikipedia, lookup

Placebo-controlled study wikipedia, lookup

Second Edition
Randomized Controlled Trials
(Experimental Studies)
What are randomized trials?
Second Edition Authors:
Lorraine K. Alexander, DrPH
Brettania Lopes, MPH
Kristen Ricchetti-Masterson, MSPH
Karin B. Yeatts, PhD, MS
Randomized trials are epidemiological
studies in which a direct comparison is
made between two or more treatment
groups, one of which serves as a control
for the other. Study subjects are
randomly allocated into the differing
treatment groups, and all groups are
followed over time to observe the effect
of the different treatments. The control
group may either be untreated (placebocontrolled) or undergo a “gold standard”
established regimen against which the
new regimen will be assessed (activecontrolled). Randomized trials provide
the most direct evidence for causality.
However, they are also fraught with a
number of additional considerations not
present for observational research.
For example, unless researchers are
genuinely uncertain about the
potential harms or benefits of a
treatment, it is unethical to assign it to
one group of people while withholding
it from others (equipoise). This limits
the types of questions that can be
answered using experimental studies.
A placebo-controlled randomized trial
might compare the effect of vitamin E
supplement in one group of
schizophrenia patients (the treatment
group) against the effects of a placebo
on a separate group of schizophrenia
patients (the control group).
An active-controlled randomized trial
might compare diabetic patients with
implanted insulin pumps against
diabetic patients who receive multiple
insulin injections (the control group).
ERIC at the UNC CH Department of Epidemiology Medical Center
Randomization avoids bias by eliminating baseline
differences in risk between treatment and control groups.
Randomization, if done properly, should make both groups
similar in terms of the distribution of risk factors,
regardless of whether these risk factors are known or
unknown (thus eliminating confounding due to both
measured and unmeasured variables). The larger the
randomized groups, the greater the probability of equal
baseline risks. However, participants in RCTs are often not
representative of the target population, which introduces
selection bias and limits generalizability.
Methods of randomization
There are different ways to randomize study participants
into treatment groups. A simple way to randomize would
be to roll a die or use a random number table to allocate
individuals into the different groups. Another way
investigators randomize study participants is through
stratified random allocation. Under this method, the
investigator first stratifies the participants by a baseline
risk factor (i.e., smoking status) then randomizes the
subjects in each stratum into either the treatment or
control group. Stratified random allocation is appropriate
when the investigator wants to be sure that a strong
external risk factor is equalized at baseline between
treatment and control groups.
Types of randomized trials
The two general types of randomized trials are clinical trials
and community trials, with randomized clinical trials being
by far the more common. A randomized clinical trial is an
experiment with patients as subjects. The goal is to find an
effective treatment for a disease or to evaluate an
intervention to prevent the progression of a disease.
Randomized clinical trials are often used to evaluate the
efficacy of new drugs against standard treatments or
against placebos, but they are also used to evaluate other
therapeutic procedures such as a new form of surgery, a
dietary regimen, or an exercise program for persons with
pre-existing disease. Most often, patients who already
PA G E 2
have some specific disease are the subjects of study in
clinical trials. However, at times, subjects who are at high
risk for a specific disease are entered into a randomized
clinical trial to assess the efficacy of a drug to prevent the
disease. For example, women with a family history of
breast cancer may be entered into a clinical trial to study
the effect of the drug tamoxifen on the prevention of
breast cancer.
A community trial is also an experiment, but differs from
clinical trials in that an entire community, rather than an
individual patient, is the unit of observation. For example,
water fluoridation was evaluated by experimentally
assigning entire communities to have their public water
supply fluoridated or not fluoridated. Units of observation
for a community trial may be a town or city, a factory or
office, a classroom or an entire school. All persons in the
same unit of observation are experimentally exposed to
the same intervention although it is not certain that all
persons in the unit will be equally exposed, e.g. that they
will drink the fluoridated water coming from their taps.
Several community trials have been conducted to
evaluate the effectiveness of mass media campaigns to
prevent heart disease by encouraging more exercise, less
use of tobacco products, and other lifestyle modifications.
Blinding or masking
Sometimes in clinical trials, participants, statisticians, and
even investigators, are made unaware of whether the
participants are part of the treatment or control group.
When only study participants are unaware of their
treatment status, but investigators and analysts are aware
of treatment status, the trial is called single-blinded.
When both the participants and the investigators are
blinded as to the treatment status of the participants the
trial is termed double-blinded. A triple-blinded trial is
when subjects, investigators, and independent
statisticians are kept unaware of subject treatment
Blinding the study participants by using placebos, or a
sham treatment, is common practice in clinical trials. The
ERIC at the UNC CH Department of Epidemiology Medical Center
placebo effect occurs when participants report a favorable
response when no treatment, but only placebo, is
administered. Another bias that is prevented by blinding of
subjects is post-randomization confounding bias where
subjects' awareness of intervention may motivate them to
be more cooperative or otherwise change their behavior.
This motivation may correlate with other risk factors for
the intended effect, thus destroying the design advantage
of randomization.
If individuals participating in a clinical trial to study the
efficacy of a new weight loss drug are aware that they
are receiving the weight loss drug, they may more
closely comply with the prescribed study diet.
Another bias that is controlled for by blinding the subjects
as to their treatment status is selection bias, or group
differences in loss to follow-up. Symptoms of disease or
side effects of the treatment may influence rates of loss to
follow-up in subjects aware of their treatment status.
Bias due to differences in reporting of symptoms, a type of
information bias, is also controlled by a double-blinded
study. Study subjects who are aware of their treatment
status may differentially report symptoms or side effects.
Likewise, staff or statisticians may differentially evaluate
subjects if they are aware of treatment status.
In a study of the effects of a new drug on severity of
migraines in which study members know their treatment
status, the treated study members may believe that the
drug will work and, therefore, report less severe
migraines. If the investigator in this study knows the
treatment status of the subjects, then that investigator
may scrutinize the severity of the migraines in treated
subjects more than that of the untreated subjects.
PA G E 3
Additional threats to the validity of a randomized trial
Limiting the analysis to compliant subjects can create bias
if compliance is correlated with other risk factors for the
treatment effect. Analyzing the results without regard to
subject compliance (called "intention-to-treat" analysis)
can help to avoid this bias. That is, subjects should be
included in the analysis whether or not they adhered to
their treatment (or control) regimen.
Suppose that in a clinical trial to look at the
relationship between diet and risk of cancer, subjects
were randomized to either a cancer-prevention diet
or to a placebo diet. Suppose again that in the
treatment group, those subjects with gastrointestinal
symptoms that are precursors of cancer, were less
compliant with their diet than subjects without
symptoms. Exclusion from the analysis of the less
compliant subjects would bias the results towards
reporting a greater effect of the cancer-prevention
diet. Only those subjects who were not at risk or who
were at low risk of cancer would be included in the
analysis. The appropriate analysis should include all
persons originally assigned to their treatment group,
whether or not they adhered to the treatments.
When noncompliant subjects are selectively excluded
from an analysis, the benefit of randomization is lost,
because unmeasured confounding factors may be
associated with the lack of compliance.
Treatment crossover
Crossover, either planned or unplanned can create biases
in experiments. In a planned crossover, group A (subjects
treated with the new drug) and group B (subjects treated
with a standard drug) would be switched to the other
treatment at the midpoint of the trial. Two of the problems
experienced with this experimental design are carryover
effects and diminished interest. Carryover effects occur
when the effects of the first drug last into the second half
ERIC at the UNC CH Department of Epidemiology Medical Center
PA G E 4
of the study when the subjects are receiving the other
treatment. Bias may also occur if there is diminished
interest or lack of compliance in the second half of the
received only the self-taught manual on meditation. Data
were collected at set intervals following the intervention to
assess the patients’ health behaviors. Studied health
behaviors included data on the patient’s diet, exercise,
and mental health. Note: this is a hypothetical example.
Unplanned crossovers occur when a clinician decides to
switch a study member from the control to the treatment
group, or vice versa, e.g. surgery vs. medical treatment for
coronary artery disease. An unplanned crossover negates
the benefit of randomization and introduces bias if
switching is related to risk of the outcome.
a) In this example, why was it ethical for the researchers to
allocate one group to receive weekly classes on meditation
practices as well as a self-taught manual on meditation
while the other group received only the self-taught manual
on meditation?
Loss to follow-up
Neither randomization nor blinding can prevent differential
loss to follow-up, or more subjects dropping out in one
treatment group than in another. Bias is introduced if the
rate of loss to follow-up is correlated with both exposure to
the treatment and exposure to other risk factors for the
Threats to validity:
Loss to follow-up
Analysis strategies to avoid bias
For purposes of analysis, study subjects should be kept in
the original randomized group, even if they were lost to
follow-up, switched to the other treatment group, or were
non-compliant (the "intention-to-treat" principle). Analysis
of any non-random subgroups threatens the validity of the
Practice Questions
Answers are located at end of this notebook.
1) Researchers conducted a multi-year ongoing
randomized controlled trial of the association between
daily meditation (such as relaxation techniques) and
health behavior among patients following a skin cancer
diagnosis. Researchers randomly allocated study
participants into 2 groups. The first patient group received
weekly classes on meditation practices as well as a selftaught manual on meditation. The second patient group
b) Which of the following may bias the analysis? Choose all
that apply.
a) Changes over time in how the health behaviors were
defined and assessed
b) Inability to blind the researchers regarding which of
the 2 meditation interventions each patient received
Lack of use of a separate untreated control group (e.g.
a group that received no meditation intervention at all)
d) Patients that are not compliant with their assigned
group (e.g. patients assigned to just the self-taught
manual but who really want to be as healthy as
possible so they show up at the weekly classes on
meditation practices)
c) In this example study, would stratified random allocation
have been useful?
Dr. Carl M. Shy, Epidemiology 160/600 Introduction to
Epidemiology for Public Health course lectures, 19942001, The University of North Carolina at Chapel Hill,
Department of Epidemiology
Rothman KJ, Greenland S. Modern Epidemiology. Second
Edition. Philadelphia: Lippincott Williams and Wilkins,
The University of North Carolina at Chapel Hill, Department
of Epidemiology Courses: Epidemiology 710,
Fundamentals of Epidemiology course lectures, 20092013, and Epidemiology 718, Epidemiologic Analysis of
Binary Data course lectures, 2009-2013.
ERIC at the UNC CH Department of Epidemiology Medical Center
PA G E 5
Answers to Practice Questions
1) In this example, researchers must be genuinely
uncertain about the potential benefits of mediation in
order for it to be ethical to assign different meditation
interventions to different groups. If the effect of
meditation on health behavior has not been extensively
studied in this study population, then that would make it
ethical to conduct this study.
b) Answer choices a, b, and d are correct. If researchers
make changes in how they assess health behaviors over
time, that would bias the results of the study. If
researchers are not able to be blinded in regard to which
patients are in which intervention group, this may lead
researchers to assess or question the study participants
about their health behaviors in different ways depending
on which group they are in. Randomization works to
eliminate baseline differences in risk between the 2
groups being compared. If some patients are not
compliant with the group they were randomized to, this
can negate the benefits of randomization. Lack of use of a
separate untreated control group does not introduce bias
into the study. The comparison groups are chosen based
on the researchers’ study hypothesis and based on what is
ethical. If the researchers wanted to study the effect of
weekly meditation classes + a self-taught manual on
mediation versus just the self-taught manual on
meditation then this is a valid study hypothesis.
The authors of the Second Edition of the ERIC Notebook would like t o acknowledge t he aut hors of
t he ERIC N ot eb ook, First Edition: Michel Ib rahim , MD, PhD, Lorraine Alexander, DrPH, Carl
Shy, MD, DrPH, Gayle Shim okura, MSPH and
Sherry Farr, GRA, Departm ent of Epidemiology
at t he Univers it y of N orth Carolina at Chapel
Hill. The First Edition of t he ERIC Notebook
was produced b y t he Educational Arm of the
Epidem iologic Research and Information Cent er at Durham, NC. The funding for the ERIC
N ot eb ook First Edit ion was provided b y t he
Departm ent of V et erans Affairs (DV A), V et erans Healt h Administ ration (V HA), Cooperative
St udies Program (CSP) to prom ot e the s t rat egic growth of t he epidem iologic capacit y of
t he DV A.
c) Stratified random allocation may have been useful.
Stratified random allocation is when the researchers first
stratify participants by a baseline risk factor and then
randomize the subjects in each stratum into the 2
comparison groups. Stratified random allocation can be
done when the researchers want to be certain that a
strong external risk factor is equalized at baseline
between the 2 comparison groups. In our hypothetical
example, researchers may have hypothesized that those
participants who had strong family support would be more
likely to learn and adopt meditation practices and more
likely to have positive health behaviors. So the researchers
could have first stratified participants based on the level of
family support they reported and then, after that,
randomized subjects into the 2 comparison groups.
ERIC at the UNC CH Department of Epidemiology Medical Center